KSampler with transparent background option

[fafayqa]

I know there is node for removing background and other such nodes, BUT they are unable removing background for example between hair or other smaller parts.

Thus it'd be great if we have KSampler where we could check some option, like for example "make transparent background", so only the main object would be created and everything else would have alpha value depending on the need, so not 0/1, but rather values in between - floats - for cases like hair or other such stuff that is normally impossible to remove background from.

This way we could TRUELY COMPOSITING from different checkpoint types (SD15, SDXL, Flux, Wan etc.) combined into one final output with seamless integration and do some other great things with it where simple remove background can not do the trick good enough!

For example we could render video of a male in Flux with true transparent background (KSampler with transparent bg option), a video of a female in WAN with true transparent background (KSampler with transparent bg option) and combine (composite) them together with another let's say static image generated with SD15, all together into one seamless smooth output.

I hope you get what I mean...

[ltdrdata]

KSampler is solely responsible for converting the input latent to the output latent using the diffusion model.

Anything related to transparency needs to be handled in the pixel space, not in the latent space.

[fafayqa]

So what node needs to make this happen - actual checkpoint model or AnimateDiff? I would like to know it so that I could go to the developer and suggest this kind of thing.

OR does that mean that if I provide images with transparent background encoded with VAE into latents and those latent fed into s KSampler, would KSampler produce output also with transparent background?

Sorry, I am no pro in this, just trying to understand the underlying process how to actually get this done.

[ltdrdata]

If you're interested in transparent images, refer to this node pack.
https://github.com/huchenlei/ComfyUI-layerdiffuse

[fafayqa]

If you're interested in transparent images, refer to this node pack. https://github.com/huchenlei/ComfyUI-layerdiffuse

thank you, it looks very interesting at first look, but the question is: does this do what I want it to do, that is video with real transparent background and not just removed background (you know, the issue with fine stuff like hair)?